Gaining Insights into Denoising by Inpainting
Authors:
Daniel Gaa,
Vassillen Chizhov,
Pascal Peter,
Joachim Weickert,
Robin Dirk Adam
Abstract:
The filling-in effect of diffusion processes is a powerful tool for various image analysis tasks such as inpainting-based compression and dense optic flow computation. For noisy data, an interesting side effect occurs: The interpolated data have higher confidence, since they average information from many noisy sources. This observation forms the basis of our denoising by inpainting (DbI) framework…
▽ More
The filling-in effect of diffusion processes is a powerful tool for various image analysis tasks such as inpainting-based compression and dense optic flow computation. For noisy data, an interesting side effect occurs: The interpolated data have higher confidence, since they average information from many noisy sources. This observation forms the basis of our denoising by inpainting (DbI) framework. It averages multiple inpainting results from different noisy subsets. Our goal is to obtain fundamental insights into key properties of DbI and its connections to existing methods. Like in inpainting-based image compression, we choose homogeneous diffusion as a very simple inpainting operator that performs well for highly optimized data. We propose several strategies to choose the location of the selected pixels. Moreover, to improve the global approximation quality further, we also allow to change the function values of the noisy pixels. In contrast to traditional denoising methods that adapt the operator to the data, our approach adapts the data to the operator. Experimentally we show that replacing homogeneous diffusion inpainting by biharmonic inpainting does not improve the reconstruction quality. This again emphasizes the importance of data adaptivity over operator adaptivity. On the foundational side, we establish deterministic and probabilistic theories with convergence estimates. In the non-adaptive 1-D case, we derive equivalence results between DbI on shifted regular grids and classical homogeneous diffusion filtering via an explicit relation between the density and the diffusion time.
△ Less
Submitted 23 September, 2023;
originally announced September 2023.
Acting Selfish for the Good of All: Contextual Bandits for Resource-Efficient Transmission of Vehicular Sensor Data
Authors:
Benjamin Sliwa,
Rick Adam,
Christian Wietfeld
Abstract:
as a novel client-based method for resource-efficient opportunistic transmission of delay-tolerant vehicular sensor data. BS-CB applies a hybrid approach which brings together all major machine learning disciplines - supervised, unsupervised, and reinforcement learning - in order to autonomously schedule vehicular sensor data transmissions with respect to the expected resource efficiency. Within a…
▽ More
as a novel client-based method for resource-efficient opportunistic transmission of delay-tolerant vehicular sensor data. BS-CB applies a hybrid approach which brings together all major machine learning disciplines - supervised, unsupervised, and reinforcement learning - in order to autonomously schedule vehicular sensor data transmissions with respect to the expected resource efficiency. Within a comprehensive real world performance evaluation in the public cellular networks of three Mobile Network Operators (MNOs), it is found that 1) The average uplink data rate is improved by 125%-195% 2) The apparently selfish goal of data rate optimization reduces the amount of occupied cell resources by 84%-89% 3) The average transmission-related power consumption can be reduced by 53%-75% 4) The price to pay is an additional buffering delay due to the opportunistic medium access strategy.
△ Less
Submitted 20 July, 2020;
originally announced July 2020.
